A method for fast prediction of large component structural deformation

By establishing a sensitivity matrix and using Bayesian inference, rapid prediction and diagnosis of structural deformation of large components were achieved. This solved the problems of online adaptive calibration and multi-source data fusion in existing technologies, improved the calculation speed and prediction accuracy, and met the needs of real-time decision-making on site.

CN123069746APending Publication Date: 2026-11-10SHANGHAI JIAOTONG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611215452.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-11
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies lack a mechanism to reduce the finite element model to a sensitivity operator that can be calculated online in real time during the docking of large components. They cannot effectively integrate multi-source data for model calibration, lack self-consistency verification and uncertainty assessment of prediction results, and are difficult to meet the real-time and accuracy requirements on site.

Method used

By establishing a finite element model, perturbing the loading points and model error parameters, solving the sensitivity matrix, establishing an influence function library, collecting data in real time, using Bayesian inference to calibrate model error parameters online, performing end-face control point prediction and force-position self-consistent diagnosis, and performing polynomial residual correction, the model can achieve rapid prediction and diagnosis.

Benefits of technology

It enables offline database building and reuse, has strong real-time online computing capabilities, can quickly complete matrix operations on ordinary industrial control computers, meets the real-time decision-making needs of docking sites, improves computing speed by 2-4 orders of magnitude, and has the ability to be embedded into existing attitude control systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069746A_ABST
    Figure CN123069746A_ABST
Patent Text Reader

Abstract

The application discloses a large component structure deformation rapid prediction method, comprising the following steps: step one, establishing a finite element model of a large component to be measured; step two, perturbing a loading point and a model error parameter, and solving a sensitivity matrix; step three, establishing a storage influence function library for calling in an online stage; step four, collecting current docking operation pose adjuster and target measurement point data in real time; step five, establishing a prediction model, and calibrating a model error parameter by using Bayesian inference; step six, performing end face control point prediction; step seven, performing force-position self-consistent diagnosis; and step eight, performing polynomial residual correction. The application has the advantages of strong real-time performance, self-adaptability, multi-source data fusion, self-consistent diagnosis capability, nonlinear residual compensation, and the like, and can improve the robustness and precision of prediction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of large component manufacturing technology, and in particular to a method for rapid prediction of structural deformation of large components. Background Technology

[0002] In ship and spacecraft construction, the docking of large components is a crucial process that determines the overall structural accuracy and quality. Taking the docking of large ship components as an example, a typical docking scenario requires the docking of two large components: a fixed section and a moving section. The fixed section is already in place, while the moving section is docked using multiple attitude adjustment and positioning devices (usually...). m×n The array of attitude adjustment devices is lifted from the ship's pier, forming a redundant parallel attitude adjustment mechanism. n-PPPS The configuration allows for six-degree-of-freedom attitude adjustment of the moving section, ensuring precise matching between the docking faces of the moving and fixed sections, thus completing the assembly. During the lifting and attitude adjustment process, the moving section undergoes structural deformation, primarily caused by its own weight leading to overall bending and localized plate indentation, load redistribution due to changes in the trolley support boundary conditions, manufacturing deviations, and stiffness non-uniformity resulting from welding residual stress. These deformations directly affect the spatial position of the docking faces; without prediction and compensation, they can lead to excessive docking clearance, additional stress from forced assembly, and even docking failure.

[0003] Currently, existing technical solutions for predicting deformation during the docking process of large components mainly fall into two categories. The first category is based on finite element simulation for full-process forward calculation. Patent document CN120805615A discloses a method for predicting stress and deformation during the assembly of ship sections. This method establishes a finite element model of the section, applies gravity, tooling constraints, and welding thermal loads, and directly solves the deformation field. The drawback of this method is that a complete nonlinear finite element calculation needs to be run again for each docking, resulting in long computation time and difficulty meeting real-time requirements on-site. Furthermore, the model parameters deviate from the actual section, and there is a lack of a mechanism for correcting the model using on-site measured data. The second category is based on comparing simulation and measured data for docking planning. Patent document CN116902168A discloses a low-stress ship section docking attitude adjustment system and its control method. This method establishes a simulated docking site and an actual docking site, adjusting the docking path by comparing target point coordinates in real time. This method results in the simulation model remaining fixed during the docking process, unable to adaptively update based on measured data; the simulation serves only as a reference benchmark, rather than a predictive tool integrated with measured data.

[0004] In summary, the common shortcomings of existing technologies include the lack of a mechanism to reduce the finite element model to a sensitivity operator that can be calculated online in real time, the lack of a statistical framework for model calibration that integrates multi-source data such as target displacement measurement and trolley force measurement, the lack of quantitative assessment of the uncertainty of prediction results, and the lack of an independent diagnostic mechanism for self-consistency verification of prediction results using trolley force information. Summary of the Invention

[0005] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is how to establish a mechanism for reducing the finite element model to a sensitivity operator that can be calculated online in real time, a statistical framework for model calibration that integrates multi-source data such as target displacement measurement and trolley force measurement, and an independent diagnostic mechanism for self-consistency verification of prediction results using positioner force information.

[0006] To achieve the above objectives, the present invention provides a method for rapid prediction of deformation of large component structures, comprising the following steps: Step 1: Establish a finite element model of the large component to be tested; Step 2: Perturb the model error parameters at each loading point and solve for the sensitivity matrix; Step 3: Establish a library of functions to store impacts for use in the online phase; Step 4: Real-time acquisition of data from the attitude adjustment positioner and target measurement points during the current docking operation; Step 5: Establish a prediction model and use Bayesian inference to calibrate the error parameters of the online calibration model; Step 6: Predict end-face control points; Step 7: Perform force-position self-consistency diagnosis; Step 8: Perform polynomial residual correction.

[0007] Furthermore, in step one, the finite element model is established using shell elements or three-dimensional solid elements, and the attitude adjustment locator is... m×n The number of target measurement points is [number]. 2l The number of target measurement points on each side of the large component is [number missing]. l Each target measurement point is equipped with a target ball, and the number of critical control points on the docking end face of the large component is [number missing]. p indivual.

[0008] Furthermore, in step two, the sensitivity matrix is ​​solved using finite element static nonlinear perturbation. The sensitivity matrix includes the first sensitivity matrix. Second sensitivity matrix Third sensitivity matrix Fourth sensitivity matrix Fifth Sensitivity Matrix .

[0009] Furthermore, in step four, the attitude adjustment and positioning device data during the current docking operation is collected in real time through the displacement encoder and force sensor, and the target ball is tracked by the laser tracker to collect the target measurement point data during the current docking operation in real time.

[0010] Furthermore, step five specifically involves establishing model error parameters. θ Prior probability distribution Likelihood function of target displacement ; Calculate model error parameters using Bayesian inference θ posterior mean and posterior covariance matrix .

[0011] Furthermore, in step five, the model error parameters... θ The prior probability distribution is solved using a multivariate Gaussian distribution, a truncated Gaussian distribution, or a log-normal distribution. The posterior probability distribution is solved using a conjugate analytical solution, variational Bayesian inference, or the expectation-maximization algorithm.

[0012] Furthermore, step six specifically involves using the posterior mean. Third sensitivity matrix and the fifth sensitivity matrix Calculate the predicted spatial coordinate vector of the end face control points. The predicted spatial coordinates of the end face control points are compared with the corresponding control point coordinates of the target installation position. The docking gap or interference at each control point is calculated to determine whether the two end faces can meet the preset docking accuracy requirements under the current deformation state.

[0013] Furthermore, step seven specifically involves using the second sensitivity matrix. and posterior mean Calculate the predicted value of the support reaction force of the attitude adjustment positioner. Design and define force-position self-consistent diagnostic indices. Calculate the difference between the measured value and the predicted value.

[0014] Further, step eight specifically involves calculating the displacement residual vector of the target measurement point. r Fitting a bivariate polynomial residual surface Extrapolate the domain of the bivariate polynomial residual surface to the coordinate positions of the end face control points. The residual corrections at each control point are obtained and then added to the prediction results from step six. .

[0015] Furthermore, the bivariate polynomial residual surface is fitted using the least squares method, radial basis function interpolation, or Kriging interpolation, and the order of the bivariate polynomial is 2nd or 3rd.

[0016] Compared with the prior art, the present invention has the following advantages: (1) Offline library creation once, reuse of all types. For multiple identical parts of the same model, only one offline finite element analysis and sensitivity matrix construction needs to be performed on the first part, and all subsequent identical parts can directly load the same influence function library; for mass production construction, the offline cost is amortized to an extremely low level.

[0017] (2) Strong real-time performance of online computation. The core computations in the online phase are matrix inversion with dimensions of 6-20 and several matrix multiplications with a maximum dimension of no more than 40-50. These computation processes consume less computer resources and take no more than 50-100 milliseconds on ordinary industrial control computers, which fully meets the real-time decision-making needs of the docking site; compared with the existing finite element forward algorithm, the speed is improved by 2-4 orders of magnitude.

[0018] (3) It can be embedded into existing attitude control systems. Deployed in the form of function libraries and calculation modules, it can be integrated into existing large component docking attitude control software as a deformation prediction and diagnosis function module, working in conjunction with existing kinematics calculation and trajectory planning modules.

[0019] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall method steps proposed in this invention, which includes an offline stage, an online stage, and a data flow direction. Figure 2 This is a schematic diagram illustrating the arrangement of large ship components, attitude adjustment and positioning array, target measurement points, and control points according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the loading point of the ship attitude adjustment and positioning device according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the arrangement of large aerospace components, attitude adjustment and positioning array, target measurement points, and control points according to an embodiment of the present invention. Figure 5 This is a schematic diagram of the finite element perturbation analysis process for constructing the sensitivity matrix in the offline stage according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the calculation process for Bayesian calibration and end-face prediction in the online stage of an embodiment of the present invention.

[0021] Among them, 201-ship moving section, 202-hull laser tracker measuring equipment, 203-hull attitude adjuster and positioner group, 204-hull target ball, 205-hull deformation key control point, 206-hull attitude adjuster and positioner guide rail, 301-large ship component, 302-ship attitude adjuster and positioner guide rail, 303-lifting loading point, 304-slipway pier, 401-large aerospace component, 402-cylinder laser tracker measuring equipment, 403-cylinder attitude adjuster and positioner group, 404-cylinder target ball, 405-cylinder deformation key control point, 406-cylinder attitude adjuster and positioner guide rail. Detailed Implementation

[0022] The preferred embodiments of the present invention are described below with reference to the accompanying drawings to make the technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0023] In the accompanying drawings, components with the same structure are indicated by the same numerical designation, and components with similar structures or functions are indicated by similar numerical designations. The dimensions and thicknesses of each component shown in the drawings are arbitrary, and the present invention does not limit the dimensions and thicknesses of each component. To make the illustrations clearer, the thickness of some components has been appropriately exaggerated in the drawings.

[0024] Currently, the technology for deformation prediction and assembly accuracy control of large components in ships, aerospace, and other industries faces technical challenges such as the inability of simulation models to be adaptively calibrated online, the lack of uncertainty quantification in prediction results, the failure to effectively integrate multi-source measurement data, and the lack of an independent diagnostic mechanism to verify the reliability of predictions.

[0025] To address the aforementioned technical problems, this invention proposes a fast prediction method for the structural deformation of large components based on sensitivity matrices and Bayesian inference, such as... Figure 1 , 5 As shown in Figure 6, this method consists of two parts: an offline phase and an online phase. The offline phase is executed once before the component under test officially enters the docking station to establish a linearized sensitivity operator library from the input to the output of the attitude adjuster. The online phase is executed during each large component docking operation, and based on the measurement data collected in real time on site, it completes model calibration, end face prediction, and self-consistent diagnosis.

[0026] Offline phase steps: Step 1: Establish the finite element model of the large component to be tested.

[0027] The finite element model is established based on the design dimensions and actual manufacturing dimensions of the component, and includes structural geometry, material properties, and m×n Information on the support position of the attitude adjuster / positioner at the bottom of the component. The positioner is arranged on both sides of the component. l Target balls, totaling2l Target measurement points. Selected on the mating end face of the component. p Several key control points are used to characterize the spatial pose of the end face. The finite element model is built using shell elements, with the mesh size not exceeding 3-5 times the wall thickness. The mesh is locally refined near the attitude adjuster support point and the target measurement point. For large components with large wall thickness, a finite element model can also be built using three-dimensional solid elements. In this case, the solution principle of the sensitivity matrix remains unchanged, only increasing the computation time in the offline stage.

[0028] Step 2: Perturb the model error parameters at each loading point and solve for the sensitivity matrix.

[0029] In the finite element model, the components are... m×n The attitude adjustment positioner is referenced to the equilibrium state of the nominal pose (which takes into account the weight of the component). Vertical force is applied sequentially at the support point of each attitude adjustment positioner. Z The unit displacement or unit force perturbation is solved by finite element static nonlinear perturbation, and the following five sensitivity matrices are recorded respectively.

[0030] First sensitivity matrix Matrix elements Indicates the first j The first unit vertical displacement caused by the attitude adjustment positioner support point i The spatial displacement response of each target measurement point can be decomposed into: X, Y, Z Three components.

[0031] Second sensitivity matrix Matrix elements Indicates the first j The first unit vertical displacement caused by the attitude adjustment positioner support point p The change in the support reaction force of the attitude adjustment positioner.

[0032] Third Sensitivity Matrix Matrix elements Indicates the first j The end face caused by the unit vertical displacement applied to the support point of the attitude adjustment positioner k A control point in a certain coordinate component ( X , Y or Z The change in ) . Here will p The three-dimensional coordinates of each control point are unfolded into a length of... 3p The vector.

[0033] Fourth sensitivity matrix Matrix element Indicates the first q Model error parameters The first when a unit change occurs i Displacement response at each target measurement point. Model error parameters. θ Used to characterize the difference between real parts and ideal finite element models. This represents the total number of model error parameters.

[0034] Fifth Sensitivity Matrix Matrix element Indicates the first q Model error parameters The end face caused by a unit change k The change of a control point on a certain coordinate component.

[0035] Among them, model error parameters θ The physical parameterization scheme includes an equivalent stiffness correction factor divided along the longitudinal direction of the component, used to characterize the local stiffness deviation from the design value caused by welding residual deformation and plate manufacturing thinning; an equivalent density correction factor, used to characterize the additional mass of equipment, accessories, pipelines, etc. not modeled in the finite element model; and foundation settlement deviation parameters at each attitude adjustment positioner support point.

[0036] The specific methods for solving the fourth and fifth sensitivity matrices are as follows: In the finite element model, a unit change is applied to the physical quantities corresponding to each model error parameter (such as the elastic modulus of a certain section of the wall panel, the material density of a certain region, and the forced displacement of a certain support point), and the displacement response of each target measurement point and end face control point is recorded through static reanalysis.

[0037] Step 3: Establish a library of functions to store impacts for use in the online phase.

[0038] The first to fifth sensitivity matrices ( , , , , ) and the target coordinate vector in the reference state End face control point coordinate vector Nominal support force vector of attitude adjuster and the nominal vertical displacement vector of the attitude adjuster. They are stored together as an impact function library file for use in the online phase.

[0039] Online phase steps: Step 4: Real-time acquisition of data from the attitude adjustment positioner and target measurement points during the current docking operation.

[0040] Step 4.1 Obtain the actual vertical displacement vector of the attitude adjustment positioner during the current docking operation. and actual support reaction vector .

[0041] The actual vertical displacement of the attitude adjustment positioner is collected in real time by the displacement encoder built into each attitude adjustment positioner; the actual support reaction force is collected in real time by the force sensor installed at the end of each attitude adjustment positioner.

[0042] Step 4.2 Obtain the current docking operation 2l Measured spatial displacement vector of each target measurement point .

[0043] The measured spatial displacement of the target measurement point is obtained by tracking the spatial coordinate changes of each target ball in real time with a laser tracker. The spatial coordinate changes are based on the target ball coordinates under the state of the large component being placed on the pier (i.e., the state before the attitude adjustment positioner is lifted up).

[0044] Step 5: Establish a prediction model and use Bayesian inference to determine the error parameters of the online calibration model.

[0045] Step 5.1 Establish model error parameters θ Prior probability distribution Likelihood function of target displacement Based on the linear superposition principle of the influence function method, the predicted value of the target displacement is... Represented as target coordinates in reference state Increment caused by trolley displacement and the increment caused by model error The sum of the three items:

[0046] Step 5.2 Based on prior probability distribution and likelihood function Calculate model error parameters using Bayesian inference θ posterior mean and posterior covariance matrix The prior probability distribution is taken as a multivariate Gaussian distribution. The prior covariance matrix The diagonal elements are set according to the physical magnitude of each model error parameter (the prior standard deviation of the stiffness correction factor is 0.05-0.15, and the prior standard deviation of the density correction factor is 0.05-0.10). The likelihood function is taken as a multivariate Gaussian distribution. Among them, the measurement noise variance Set according to the nominal measurement accuracy of the laser tracker.

[0047] Under the conjugate prior assumption, the posterior mean and posterior covariance matrix have the following analytical expressions:

[0048] The advantages of using the conjugate analytical solution are: the computation time is controllable, making it suitable for real-time use in the docking field; and it has good numerical stability and does not have sampling convergence problems.

[0049] When engineering experience shows that the distribution of certain parameters has obvious boundary constraints, the prior probability distribution of the model error parameters adopts a truncated Gaussian distribution or a log-normal distribution, and the posterior distribution is solved by variational Bayesian inference or expectation-maximization algorithm.

[0050] Step 6: Predict the end face control points.

[0051] Step 6.1 Using the posterior mean Third sensitivity matrix and the fifth sensitivity matrix Calculate the end face p Predicted spatial coordinate vector of each control point :

[0052] Step 6.2 Face the end p The predicted spatial coordinates of each control point are compared with the coordinates of the corresponding control points at the target installation position (point cloud data or theoretical design coordinates obtained in advance by a laser scanner), and the docking gap or interference at each control point is calculated to determine whether the two end faces can meet the preset docking accuracy requirements under the current deformation state.

[0053] Step 7: Perform force-position self-consistency diagnosis.

[0054] Step 7.1 Using the second sensitivity matrix and posterior mean Calculate the predicted value of the support reaction force of the attitude adjustment positioner. :

[0055] in, Model error parameters θ The sensitivity matrix for the support reaction force of the attitude adjuster positioner is calculated and stored in the influence function library along with the first to fifth sensitivity matrices during the offline phase.

[0056] Step 7.2 Design and define force-potential self-consistent diagnostic indices It is used to calculate the difference between the measured value and the predicted value.

[0057]

[0058] in, Let be the measurement noise covariance matrix of the force sensor in the attitude adjuster. The physical meaning of this index is: assuming the model is correct, the expected value of the normalized sum of squared force residuals is 1. When When the prediction exceeds a preset threshold, it is determined that the current prediction result contains unmodeled systematic errors, triggering model recalibration or manual review. Possible causes of systematic errors include contact nonlinearity, localized gaps between the support point and the bottom of the component, or discrepancies between the finite element model and the actual component. θ Differences in parameterization range. The information provided by the force sensor and the target displacement measurement information are mechanically independent, thus this diagnostic provides independent cross-validation of the prediction reliability.

[0059] Step 8: Perform polynomial residual correction.

[0060] Step 8.1 Calculate the displacement residual vector of the target measurement point r .

[0061]

[0062] The residual reflects the residual deformation components that the linear sensitivity model and model error parameterization failed to capture. The main sources include nonlinear effects such as local lattice bending and cross-sectional distortion.

[0063] Step 8.2 Fit the bivariate polynomial residual surface.

[0064] Displacement residual vector r According to each side l The target points are grouped into groups, and the longitudinal coordinates of each target measurement point on the side of the component under test are used as the basis for the measurement. x and vertical coordinates y Using as the independent variable, fit a bivariate polynomial residual surface. The bivariate polynomial has an order of 2 or 3, and is fitted using the least squares method. For a 2nd-order polynomial, there are 6 undetermined coefficients. For a third-order polynomial, there are 10 undetermined coefficients. The fitting process uses each side... l The residual data of each target point is much larger than the undetermined coefficients of the polynomial, and the fitting has good numerical stability.

[0065] For bivariate polynomial residual surfaces, when the target points are unevenly distributed or the end face shape is complex, radial basis function interpolation or kriging interpolation can be used for fitting to better handle irregular sampling and extrapolation problems.

[0066] Step 8.3 Extrapolates the domain of the bivariate polynomial residual surface to the end face. p Coordinates of each control point The residual corrections at each control point are obtained and then added to the prediction results from step six. :

[0067] This polynomial correction step only slightly modifies the prediction results and does not affect the main part of the prediction results obtained in step eight, but it can effectively capture local nonlinear effects that are difficult to model under the linear sensitivity framework.

[0068] The implementation and realization of the technical solution of the present invention will be further described below through specific embodiments. It should be noted that the implementation and realization of the technical solution of the present invention are not limited to the following embodiments.

[0069] Example 1

[0070] This embodiment illustrates the docking of large components in shipbuilding, such as... Figure 2 The diagram shows the layout of the docking system for large ship components, including the ship's moving section 201, the hull laser tracker measuring equipment 202, the hull attitude adjustment and positioning device group 203 arranged in m rows and n columns, the hull target balls 204 pasted on the hull surface (1 on each side), the p hull deformation key control points 205 on the docking end face, and the hull attitude adjustment and positioning device guide rails 206.

[0071] Among them, the large component is the closed-type hull moving section 201, with a rectangular closed outer perimeter, an internal mid-deck, and several rows of longitudinal stiffeners to increase structural rigidity. The bottom of the component is... m×n = 4×4 = 16 attitude adjusters array support. Each attitude adjuster has forward and backward degrees of freedom along the guide rail direction, as well as end POGO column. X, Y, Z With three degrees of translational freedom, 16 positioners together form a 16-PPPS redundant parallel attitude adjustment mechanism. Positioners are arranged on both sides of the component. l =18 laser tracker hull target spheres, 204 in total, 2l = 36 target measurement points. Target points on each side are arranged in 3 rows ( X (direction) × 6 columns ( Z The surfaces are evenly distributed, without touching the side edges. Selected on the mating end faces... p = 20 deformation key control points, evenly distributed along the end face contour, are used to characterize the spatial pose and shape of the end face.

[0072] In this embodiment, the offline phase is executed according to the following process, such as Figure 5 As shown.

[0073] First, a finite element model was established. The moving section was meshed using quadrilateral shell elements with a mesh size of 100 mm. Mesh refinement was applied within a 500 mm radius around the attitude adjuster support point and a 300 mm radius around the target measurement point. The finite element model has a total of 117,437 nodes and 704,622 degrees of freedom.

[0074] Secondly, the reference state is solved. In the finite element model, vertical constraints are applied to the 16 attitude adjustment positioner support points, and a total gravity load is applied to the component. To prevent rigid body displacement, three non-collinear nodes not on the same straight line are selected on the model to apply minimum additional constraints. The static equilibrium is solved to obtain the support reaction forces of each attitude adjustment positioner under the reference state. ), spatial coordinates of each target measurement point ( ), spatial coordinates of p control points on the end face ( ).

[0075] Then, the sensitivity matrix is ​​solved. Based on the finite element model that converges to the reference state, linear perturbation analysis is performed on the ... j One attitude adjustment positioner support point ( j = 1, 2, ..., 16 ), at this node Z To apply δ z A forced displacement of 1.0 mm was achieved while maintaining the remaining support points. Z With constraints unchanged, solve for the changes in the perturbation displacement field and reaction force field, and record the displacement changes at the target measurement points. ), the change in reaction force at the support point ( ), Changes in the coordinate components of the end face control points ( ).

[0076] At the same time, the model error parameters Apply unit changes to each parameter, solve for the equilibrium state changes under reference state boundary conditions, and record the target measurement points ( ) and end face control points ( The displacement response of ). In this embodiment, the model error parameters It includes 5 stiffness correction factors and 1 density correction factor, and the prior probability distribution of the model error parameters adopts a multivariate Gaussian distribution.

[0077] In this example, the online phase of each docking operation is executed according to the following process: Figure 6 As shown.

[0078] First, data acquisition is performed. After the ship's moving section 201 is lifted from the slipway pier 304 to the predetermined height by 16 attitude adjustment and positioning devices, the displacement encoders of each attitude adjustment and positioning device read back the vertical displacement. Force sensor reads back support reaction force Simultaneously, the laser tracker measured the real-time spatial coordinates of 36 target spheres 204 on both sides of the ship's hull, and compared them with the reference coordinates of the landing state to obtain the measured displacement vector. .

[0079] Next, perform Bayesian calibration. Load the influence function library and proceed step by step.5.1 Established prediction model Set the standard deviation of the measurement noise σ = 0.02 mm, and calculate the posterior mean. and posterior covariance It can be completed within 30ms on a standard industrial control computer.

[0080] Then, end-face prediction is performed, following the steps. 6.1 calculate p = Predicted spatial coordinates of 20 end face control points .

[0081] In addition, you can follow the steps 7.1-7.2 calculate Indicators. If If < 3.0, the prediction result is deemed reliable, and the end face prediction result and docking clearance report are output; if If the value is ≥3.0, the system will indicate that there may be abnormal contact or significant model deviations that exceed the parameterization range, and recommend that the user or operator perform a manual review.

[0082] In addition, you can follow the steps 8.1-8.2 Calculate the displacement residual vector of 36 target points r The residuals of the 18 target points on each of the left and right sides are respectively divided into ( x, z The coordinates are fitted using the least squares method to a bivariate polynomial residual surface. The polynomial order is second or third, and the coefficients are obtained. The polynomial is then extrapolated to the positions of 20 end-face control points, following the steps... 8.3 The formula is superimposed and corrected, and the corrected predicted coordinates of the control points are output.

[0083] Example 2

[0084] This embodiment describes the docking of large components in shipbuilding; other aspects are the same as in Embodiment 1. Figure 3 The diagram shows the loading point of the attitude adjustment and positioning device in the docking system for large ship components, including the large ship component 301, the ship attitude adjustment and positioning device guide rail 302, the lifting loading point 303 where the attitude adjustment and positioning device contacts the component, and the slipway pier 304.

[0085] Example 3

[0086] This embodiment describes the docking of large components in spacecraft manufacturing; other aspects are the same as in Embodiment 1. Figure 4 The diagram shows the docking system layout of a large spacecraft component, including a large spacecraft component 401, a cylinder laser tracker measuring device 402, a cylinder attitude adjuster and positioner group 403, cylinder target balls 404 pasted on the cylinder surface (one on each side), p cylinder deformation key control points 405 set on the docking end face, and cylinder attitude adjuster and positioner guide rails 406.

[0087] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for rapid prediction of deformation of large component structures, characterized in that, Includes the following steps: Step 1: Establish a finite element model of the large component to be tested; Step 2: Perturb the model error parameters at each loading point and solve for the sensitivity matrix; Step 3: Establish a library of functions to store impacts for use in the online phase; Step 4: Real-time acquisition of data from the attitude adjustment positioner and target measurement points during the current docking operation; Step 5: Establish a prediction model and use Bayesian inference to calibrate the error parameters of the online calibration model; Step 6: Predict end-face control points; Step 7: Perform force-position self-consistency diagnosis; Step 8: Perform polynomial residual correction.

2. The method for rapid prediction of deformation of large component structures as described in claim 1, wherein in step one, the finite element model is established using shell elements or three-dimensional solid elements, and the attitude adjustment locator is... m×n The number of target measurement points is [number]. 2l The number of target measurement points on each side of the large component is [number]. l Each target measurement point is equipped with a target ball, and the number of key control points on the docking end face of the large component is [number missing]. p indivual.

3. The rapid prediction method for deformation of large component structures as described in claim 2, wherein in step two, the sensitivity matrix is ​​solved by finite element static nonlinear perturbation, and the sensitivity matrix includes a first sensitivity matrix. Second sensitivity matrix Third sensitivity matrix Fourth sensitivity matrix Fifth Sensitivity Matrix .

4. In the rapid prediction method for deformation of large component structures as described in claim 2, in step four, the attitude adjustment and positioning device data in the current docking operation is collected in real time by a displacement encoder and a force sensor, and the target measurement point data in the current docking operation is collected in real time by a laser tracker tracking the target ball.

5. The rapid prediction method for deformation of large component structures as described in claim 3, wherein step five specifically involves establishing model error parameters. θ Prior probability distribution Likelihood function of target displacement The model error parameters are calculated using Bayesian inference. θ posterior mean and posterior covariance matrix .

6. The method for rapid prediction of deformation of large component structures as described in claim 5, wherein in step five, the model error parameter... θ The prior probability distribution is solved using a multivariate Gaussian distribution, a truncated Gaussian distribution, or a log-normal distribution, and the posterior probability distribution is solved using a conjugate analytical solution, variational Bayesian inference, or the expectation-maximization algorithm.

7. The method for rapid prediction of deformation of large component structures as described in claim 5, wherein step six specifically involves using the posterior mean... The third sensitivity matrix and the fifth sensitivity matrix Calculate the predicted spatial coordinate vector of the end face control point. The predicted spatial coordinates of the end face control points are compared with the coordinates of the corresponding control points of the target installation position. The docking gap or interference at each control point is calculated to determine whether the two end faces can meet the preset docking accuracy requirements under the current deformation state.

8. The method for rapid prediction of deformation of large component structures as described in claim 7, wherein step seven specifically involves using the second sensitivity matrix... and the posterior mean Calculate the predicted value of the support reaction force of the attitude adjuster positioner. Design and define force-position self-consistent diagnostic indices. Calculate the difference between the measured value and the predicted value.

9. The method for rapid prediction of deformation of large component structures as described in claim 1, wherein step eight specifically involves calculating the displacement residual vector of the target measurement point. r Fitting a bivariate polynomial residual surface Extrapolate the domain of the bivariate polynomial residual surface to the coordinate position of the end face control point. The residual corrections at each control point are obtained and then added to the prediction results from step six. .

10. The method for rapid prediction of deformation of large component structures as described in claim 9, wherein the bivariate polynomial residual surface is fitted using the least squares method, radial basis function interpolation method, or Kriging interpolation method, and the order of the bivariate polynomial is 2nd or 3rd order.

Citation Information

Patent Citations

  • Low-stress ship block butt joint attitude adjusting system and control method thereof

    CN116902168A

  • Prediction method for stress deformation in ship block closure process

    CN120805615A